Market integration, systemic risk and diagnostic tests in large mixed panels

Abstract

This study investigates an AR (autoregressive)-filtered version of several conventional diagnostic tests for cross-sectional dependence in large mixed panels when both N and T are large, including the adjusted Lagrangian Multiplier test (LM), the cross-section dependence test (CD), and the Schott test. We show that conventional tests of cross-sectional dependence based on Pearson correlation coefficients could diverge if the components are not all I(0) processes and the modified tests possess the asymptotical normality property. The distinctive feature of these new tests is their ease of implementation, even though the exact time series properties of each component of a mixed panel are unknown or unobservable in practice. Simulations show that the AR-filtered version of the CD test (CDAR ) performs well relative to the other testing procedures in the finite sample and computation time, especially for those cases with a large cross-sectional dimension. Given the good statistical properties of CDAR test, we also propose to use it as an early warning indicator for market risk or crisis.